Unclear interaction between `--modified-bases` `model` and automatic model selection

[awgymer]

Issue Report

Please describe the issue:

The purpose of, and interaction between, the --modified-bases and model params are unclear, particularly as it pertains to the automatic model selection.

Consider the following scenarios:

• dna_r10.4.1_e8.2_400bps_hac@v4.2.0 downloaded with dorado download and passed as a model path to dorado basecaller: basecalling performed no methylation info
• dna_r10.4.1_e8.2_400bps_hac@v4.2.0 downloaded with dorado download and passed as a model path to dorado basecaller and --modified-bases 5mCG_5hmCG supplied: dorado_basecaller recognises that the supplied model has no mehtylation and so the logs show it downloads the appropriate one downloading dna_r10.4.1_e8.2_400bps_hac@v4.2.0_5mCG_5hmCG@v2 with httplib. Basecalling performed and methylation information included
• dorado basecaller called with a model param of hac@v4.2.0 and --modified-bases 5mCG_5hmCG: dorado basecaller downloads the model without methylation info downloading dna_r10.4.1_e8.2_400bps_hac@v4.2.0 with httplib. Basecalling performed with no methylation info added.

This inconsistent behaviour regarding the modified-bases flag and whether or not a modified model is downloaded is unclear from any documentation, and indeed it seems that with automatic model selection that parameter is just sliently ignored?

Run environment:

• Dorado version: 0.5.3
• Operating system: ubuntu
• Hardware (CPUs, Memory, GPUs): nvidia a100
• Source data type (e.g., pod5 or fast5 - please note we always recommend converting to pod5 for optimal basecalling performance): pod5

[tijyojwad]

Hi @awgymer - this was a bug in 0.5.3 that's now fixed in 0.6.0. Running hac@v4.2.0 with --modified-bases should also download the modified bases model and run basecalling with methylation.

[awgymer]

I see. I will try to update to 0.6.0 ASAP

But is there a reason to have multiple conflicting ways of specifying the methylation parameter?

It feels like there is some weird interactions possible e.g. what would the expected behaviour be here?

dorado basecaller hac@v4.3.0,5mCG_5hmCG,6mA pod5_files --modified-bases 5mCG_5hmCG

[tijyojwad]

$ dorado basecaller hac@v4.3.0,5mCG_5hmCG,6mA pod5 --modified-bases 5mCG_5hmCG
[2024-04-04 09:30:18.587] [info] Running: "basecaller" "hac@v4.3.0,5mCG_5hmCG,6mA" "pod5" "--modified-bases" "5mCG_5hmCG"
[2024-04-04 09:30:18.588] [error] Only one of --modified-bases, --modified-bases-models, or modified models set via models argument can be used at once

[awgymer]

Okay so that's good. Thanks.